Midjourney Prompt Architect
Generate production-quality prompts for AI image generation tools.
Prompt Architecture
Structure (in order)
CODEBLOCK0
Component Details
Subject & Action
- - Be specific: "a lone astronaut floating" not "a person in space"
- Include age, clothing, expression when relevant
- Use participial phrases for dynamic poses: "reaching toward a glowing orb"
Setting & Environment
- - Time of day, weather, season
- Specific location details: "abandoned cathedral with stained glass"
- Atmospheric elements: "mist curling between pillars"
Style
Reference established art movements, artists, or media:
- - INLINECODE0
- INLINECODE1
- INLINECODE2
- INLINECODE3
- INLINECODE4
Lighting (critical for quality)
CODEBLOCK1
Camera & Composition
CODEBLOCK2
Color Palette
CODEBLOCK3
Midjourney Parameters
CODEBLOCK4
Workflow
1. Understand the Vision
Ask clarifying questions if the request is vague:
- - What mood or emotion?
- Any reference images or artists?
- Realistic, stylized, or abstract?
- Intended use (print, social, web, concept art)?
2. Draft → Refine → Test Cycle
- 1. Write initial prompt from the architecture above
- Evaluate against common failure modes (see below)
- Add Midjourney-specific parameters
- Provide 3-4 variations with different creative directions
- Suggest negative prompts to avoid common artifacts
3. Batch Generation Template
For repeated themes, create a parameterized template:
CODEBLOCK5
Common Failure Modes & Fixes
| Problem | Fix |
|---|
| Text appears in image | Add INLINECODE5 |
Faces look distorted | Reduce
--s, add
detailed face, symmetrical |
Multiple subjects merge | Specify
two separate figures, clear spacing |
Too busy/cluttered | Reduce subject count, add
minimalist, clean composition |
Looks like AI art | Reference specific artist/medium, avoid
AI art, digital illustration |
Inconsistent style across batch | Fix seed with
--seed [value], keep parameters identical |
Watermarks | Add
--no watermark, logo, signature |
Prompt Examples
Cinematic Portrait
CODEBLOCK6
Architectural Concept
CODEBLOCK7
Fantasy Illustration
CODEBLOCK8
Style Quick Reference
| Want This | Use This |
|---|
| Photorealistic | INLINECODE13 |
| Oil painting |
oil painting, impasto technique, visible brushstrokes |
| Watercolor |
watercolor wash, wet-on-wet, soft edges, paper texture |
| 3D render |
3D render, Unreal Engine 5, octane render, ray tracing |
| Pencil sketch |
graphite pencil sketch, cross-hatching, on textured paper |
| Pixel art |
16-bit pixel art, limited color palette, retro game aesthetic |
| Anime |
anime style, cel shading, clean lineart, vibrant colors |
Cross-Tool Compatibility
These prompts work with minor adjustments:
- - Stable Diffusion: Remove
-- parameters, add to negative prompt box - DALL-E 3: Use natural language, drop camera specs
- Flux: Keep descriptive, reduce technical jargon
Midjourney 提示词架构师
为AI图像生成工具生成专业级提示词。
提示词架构
结构(按顺序)
[主体] + [动作/姿势] + [环境/场景] + [风格] + [光线] + [镜头/构图] + [色调] + [技术参数] + [负面提示词]
组件详解
主体与动作
- - 具体描述:一名孤独漂浮的宇航员而非太空中的一个人
- 必要时包含年龄、服装、表情
- 使用分词短语描述动态姿势:伸向一个发光球体
环境与场景
- - 时间、天气、季节
- 具体地点细节:带有彩色玻璃的废弃大教堂
- 氛围元素:柱子间缭绕的薄雾
风格
参考已确立的艺术流派、艺术家或媒介:
- - J.M.W.特纳风格的油画
- 吉卜力工作室动画赛璐珞片
- 微距摄影,焦点堆叠
- 粗野主义建筑,混凝土,原始质感
- 浮世绘木版画,葛饰北斋
光线(对质量至关重要)
黄金时刻,体积光,轮廓光
生物发光,湿漉漉路面上的霓虹倒影
明暗对比,伦勃朗布光
影棚布光,柔光箱,三点布光
正午烈日,深邃阴影
镜头与构图
35mm镜头,f/1.4,浅景深
无人机航拍,200米高度,鸟瞰视角
荷兰角,低角度仰拍
广角畸变,鱼眼镜头
移轴微缩效果
色调
单色蓝,低饱和度
互补橙与青绿
柔和大地色,赭石与鼠尾草绿
高对比度,饱和原色
粉彩,柔和粉红与淡紫
Midjourney参数
--ar 16:9 # 宽高比
--s 750 # 风格化(0-1000)
--c 50 # 混乱度(0-100)
--q 2 # 质量
--v 6.1 # 模型版本
--no 文字, 水印 # 负面提示词
--tile # 无缝拼接
--iw 2.0 # 图像权重(用于图像参考)
工作流程
1. 理解愿景
如果需求模糊,提出澄清性问题:
- - 想要什么情绪或情感?
- 有参考图像或艺术家吗?
- 写实、风格化还是抽象?
- 预期用途(印刷、社交媒体、网页、概念艺术)?
2. 草稿→优化→测试循环
- 1. 根据上述架构编写初始提示词
- 对照常见失败模式进行评估(见下文)
- 添加Midjourney特定参数
- 提供3-4个不同创意方向的变体
- 建议负面提示词以避免常见瑕疵
3. 批量生成模板
对于重复主题,创建参数化模板:
[主体]以[风格]呈现,[场景],[光线],[镜头],--ar [比例] --s [风格值] --v 6.1 --no [负面词]
常见失败模式及修复
| 问题 | 修复 |
|---|
| 图像中出现文字 | 添加--no 文字, 字母, 排版, 水印 |
| 面部扭曲 |
降低--s,添加detailed face, symmetrical |
| 多个主体融合 | 指定two separate figures, clear spacing |
| 过于杂乱 | 减少主体数量,添加minimalist, clean composition |
| 看起来像AI艺术 | 引用特定艺术家/媒介,避免AI art, digital illustration |
| 批次间风格不一致 | 使用--seed [值]固定种子,保持参数一致 |
| 水印 | 添加--no watermark, logo, signature |
提示词示例
电影感肖像
一位饱经风霜的渔夫站在黎明时分的木码头上,脸庞被温暖的金色光芒照亮,胡须上结着盐霜,手持手工编织的渔网,平静的海面倒映着橙色的天空,哈苏500C拍摄,80mm镜头,f/2.8,浅景深,柯达Portra 400胶片模拟 --ar 4:5 --s 800 --v 6.1
建筑概念
建在悬崖边的未来主义粗野主义图书馆,悬臂式混凝土书架上堆满书籍,落地玻璃窗俯瞰峡湾,裂缝间长满青苔,斑驳光线透过几何天窗洒落,建筑摄影,广角镜头,蓝色时刻光线 --ar 21:9 --s 600 --v 6.1 --no 人物, 汽车, 现代标识
奇幻插画
一只长着彩虹色皮毛的小狐仙坐在新月上,阅读着一卷展开成星河的书卷,周围环绕着漂浮的灯笼,日式水墨画遇见新艺术风格,深靛蓝背景上的金箔点缀 --ar 1:1 --s 900 --c 30 --v 6.1
风格快速参考
| 想要这种效果 | 使用这些关键词 |
|---|
| 照片级写实 | photograph, shot on [相机], [胶片], f/[光圈] |
| 油画 |
oil painting, impasto technique, visible brushstrokes |
| 水彩 | watercolor wash, wet-on-wet, soft edges, paper texture |
| 3D渲染 | 3D render, Unreal Engine 5, octane render, ray tracing |
| 铅笔素描 | graphite pencil sketch, cross-hatching, on textured paper |
| 像素艺术 | 16-bit pixel art, limited color palette, retro game aesthetic |
| 动漫 | anime style, cel shading, clean lineart, vibrant colors |
跨工具兼容性
这些提示词经过微小调整即可使用:
- - Stable Diffusion:移除--参数,添加到负面提示词框
- DALL-E 3:使用自然语言,去掉相机参数
- Flux:保持描述性,减少技术术语